As a researcher deeply invested in the advancement of electric vehicle technology, I have long recognized the critical role that accurate State of Charge (SOC) estimation plays for li ion battery systems. The SOC, essentially the battery’s equivalent of a fuel gauge, directly influences battery management, safety, longevity, and overall vehicle performance. Inaccurate SOC readings can lead to overcharge, over-discharge, reduced efficiency, and even hazardous conditions. Consequently, my work has focused on refining one of the most widely implemented SOC estimation methods—the ampere-hour (Ah) integral method—by grounding it in the fundamental electrochemistry of the li ion battery.
The traditional Ah integral method, while simple and computationally efficient, suffers from significant drawbacks. Its core formula is given by:
$$S = S_0 + \frac{1}{C_N} \int_{t_0}^{t} \eta(t) \cdot I(t) dt$$
Here, \(S\) is the current SOC, \(S_0\) is the initial SOC, \(C_N\) is the nominal capacity, \(\eta\) is the coulombic efficiency, and \(I(t)\) is the time-varying current. The primary challenge lies in the parameter \(\eta\) and the effective capacity \(C_{total}(t)\), which are highly nonlinear functions of operating current and temperature. In practice, empirical formulas are often used to approximate \(C_{total}(t)\), but these fits can introduce substantial errors, especially under dynamic loads or varying thermal conditions. Furthermore, during rapid current transitions, the traditional method can produce abrupt, non-physical jumps in the estimated SOC, which misrepresents the true state of the li ion battery.
To address these limitations, I proposed an improved Ah integral method that dynamically calculates the li ion battery’s actual available capacity based on the lithium concentration within the electrode’s active material particles. This approach leverages a single-particle model (SPM) for lithium diffusion, which, while simplified, captures the essential mass transport limitations that govern capacity under load. The fundamental premise is that the total charge that can be delivered by a li ion battery is directly proportional to the amount of lithium ions that can be extracted from (or inserted into) the anode particles before a cutoff voltage is reached.
The improved estimation formula is expressed as:
$$S = S_0 + \frac{1}{C_{total}(t)} \int_{t_0}^{t} I(t) dt$$
where the total effective capacity \(C_{total}(t)\) is no longer a constant or a simple empirical fit. Instead, it is the sum of the capacity already discharged, \(C_{dis}(t)\), and the remaining available capacity, \(C_{ava}(t)\):
$$C_{total}(t) = C_{dis}(t) + C_{ava}(t)$$
$$C_{dis}(t) = \int_{t_0}^{t} I(t) dt$$
The available capacity is derived from the lithium inventory in the anode. For a graphite-based anode, common in many commercial li ion battery cells, it is calculated as:
$$C_{ava}(t) = \begin{cases}
26.8 L A \varepsilon_s [c_s(r_s, t) – c_{s\_cut}], & \text{if } c_s(r_s, t) = \min(c_s(r, t)) \\
26.8 L A \varepsilon_s [-c_s(r_s, t) – c_{s\_cut}], & \text{if } c_s(r_s, t) > \min(c_s(r, t))
\end{cases}$$
In these equations, \(L\) is the electrode coating thickness, \(A\) is the coating area, \(\varepsilon_s\) is the volume fraction of active material, \(r_s\) is the particle radius, \(c_s(r,t)\) is the spatially-varying lithium concentration within a spherical particle, \(\overline{c_s}\) is the average concentration, and \(c_{s\_cut}\) is the lithium concentration at the anode particle surface when the cell voltage reaches the discharge cutoff voltage. The factor 26.8 converts molar concentration to ampere-hours (based on Faraday’s constant).
The heart of this improved method is solving for the lithium concentration profile \(c_s(r,t)\) within the active material particle. This is governed by Fick’s second law of diffusion in spherical coordinates:
$$\frac{\partial c_s}{\partial t} = \frac{D_s}{r^2} \frac{\partial}{\partial r} \left( r^2 \frac{\partial c_s}{\partial r} \right)$$
Subject to the boundary conditions:
$$-D_s \left. \frac{\partial c_s}{\partial r} \right|_{r=0} = 0$$
$$-D_s \left. \frac{\partial c_s}{\partial r} \right|_{r=r_s} = -\frac{I}{a F L A}$$
Here, \(D_s\) is the solid-phase lithium diffusion coefficient, \(a\) is the specific surface area, and \(F\) is Faraday’s constant. Crucially, \(D_s\) is temperature-dependent, modeled using an Arrhenius-type equation to account for thermal effects on the li ion battery’s kinetics:
$$D_s = 1.4523 \times 10^{-13} \times \exp\left[ -\frac{20000}{8.314} \left( \frac{1}{T} – \frac{1}{T_{ref}} \right) \right]$$
with \(T_{ref} = 298.15 \, K\). This temperature dependence is vital for accurate SOC estimation across the operational range of a li ion battery.
To implement this model for online SOC monitoring, I developed a solver program within the LabVIEW graphical programming environment. The program architecture consisted of two parallel modules: one for the traditional Ah integral method (using a fitted capacity model) and one for the improved diffusion-based method. A key component was a Virtual Instrument Software Architecture (VISA) communication interface to acquire real-time data from battery testing equipment. This setup allowed for simultaneous online estimation and comparison.
The experimental platform centered on a commercial soft-pack li ion battery with a graphite anode and a Nickel Cobalt Manganese (NCM) oxide cathode. Its key parameters are summarized in the table below, which are essential inputs for the diffusion model.
| Parameter | Symbol | Value | Unit |
|---|---|---|---|
| Nominal Capacity | \(C_N\) | 26 | Ah |
| Anode Coating Thickness | \(L\) | 50 | μm |
| Electrode Area | \(A\) | 1.1314 | m² |
| Cutoff Li Concentration | \(c_{s\_cut}\) | 1255 | mol/m³ |
| Particle Radius | \(r_s\) | 6.0 | μm |
| Active Material Volume Fraction | \(\varepsilon_s\) | 62.7 | % |
| Open Circuit Voltage | \(U_N\) | 3.65 | V |
For the traditional method’s capacity model, discharge tests were conducted at various temperatures and constant currents (C-rates) to fit an empirical function. The measured capacities for the li ion battery under test are shown below, highlighting the nonlinear dependence on both current and temperature.
| Discharge C-rate | 5°C Measured (Ah) | Room Temp Measured (Ah) | 30°C Measured (Ah) | Fitted Capacity (Ah)* |
|---|---|---|---|---|
| 0.5C | 24.6 | 26.2 | 26.7 | 24.2 / 26.2 / 27.0 |
| 1.0C | 23.9 | 25.8 | 26.4 | 23.8 / 25.7 / 26.5 |
| 1.5C | 23.5 | 25.6 | 26.1 | 23.6 / 25.5 / 26.2 |
| 2.0C | 23.3 | 25.5 | 25.9 | 23.5 / 25.3 / 25.9 |
| 2.5C | 23.2 | 25.4 | 25.8 | 23.4 / 25.2 / 25.7 |
*Fitted values are listed for 5°C / Room Temp / 30°C respectively, based on the empirical model.
The empirical model for total capacity \(C_{total}(t)\) in the traditional method took the form:
$$C_{total}(t) = (p_1 + p_2 R + p_3 R^2 + p_4 R^3) \cdot (p_5 – p_6 \cdot e^{-p_7 T})$$
where \(R\) is the discharge C-rate (\(|I|/I_{1C}\)) and \(p_1\) to \(p_7\) are fitted parameters. While this model showed reasonable correlation (\(R^2 > 0.98\)), its inherent limitation is the assumption of a static relationship, which fails to capture the dynamic redistribution of lithium within particles during transient operations—a phenomenon central to the performance of a li ion battery.
Three distinct test profiles were used to evaluate and compare the SOC estimation methods: Constant Current-Constant Voltage (CC-CV) charging, a variable current pulse discharge, and an Extra-Urban Driving Cycle (EUDC) discharge. The latter two were tested at both room temperature (22-27°C) and a low temperature of 10°C to assess thermal robustness. Since SOC cannot be measured directly, estimation accuracy was evaluated by comparing the time \(t_{SOC}\) when the estimated SOC reached 0% (or 100% for charging) with the actual time \(t_{cut}\) when the li ion battery terminal voltage hit the predefined cutoff voltage (2.8 V for discharge, 4.2 V for charge). The error metric was:
$$\varepsilon = \left| \frac{t_{SOC} – t_{cut}}{t_{cut}} \right| \times 100\%$$
Detailed Analysis of Results and Method Performance
In the CC-CV charge test, both methods performed identically and accurately, estimating a final SOC of 99.2%. This was expected because during charging, the effective capacity for both algorithms was set to the nominal capacity \(C_N\). This test primarily served to initialize the SOC for discharge tests and confirm basic functionality. The real differentiation began during dynamic discharge.
For the variable current pulse discharge test, the current profile consisted of repeated sequences of high discharge, rest, low discharge, and rest periods. The results were telling. At room temperature, the traditional Ah integral method yielded an SOC estimation error \(\varepsilon\) of 0.78%, while the improved diffusion-based method achieved a remarkable error of only 0.07%. At 10°C, the gap widened: 1.01% error for the traditional method versus 0.04% for the improved method. More importantly, the temporal evolution of the SOC estimates revealed a critical advantage. The traditional method exhibited sharp “jumps” or “dips” in the SOC trace whenever the load current changed abruptly. For instance, when transitioning from a high discharge current to a rest period, the estimated SOC would show a sudden increase, which is physically counterintuitive. Conversely, the improved method produced a smooth, monotonic decline in SOC with only minor, physically explainable fluctuations. This smoothness stems from the model’s inherent accounting for lithium concentration gradients. When a high current stops, lithium continues to diffuse from the particle interior to the surface, gradually increasing the surface concentration and thus the measurable voltage and available capacity. The improved method’s \(C_{ava}(t)\) reflects this recovery, preventing artificial SOC spikes.
The EUDC profile, designed to mimic real-world driving conditions, presented a more complex and irregular current demand. Here, the superiority of the physics-based approach became even more evident. At room temperature, the traditional method’s error was 1.71%, compared to 1.11% for the improved method. At 10°C, the errors were 1.89% and 0.88%, respectively. The improved method consistently cut the error nearly in half under these realistic, dynamic conditions. The table below summarizes the estimation errors across all test cases, clearly demonstrating the enhanced accuracy provided by integrating the single-particle li ion battery model.
| Test Profile | Temperature | Traditional Ah Error (\(\varepsilon\)) | Improved Diffusion-Based Error (\(\varepsilon\)) |
|---|---|---|---|
| CC-CV Charge | Room Temp | ~0.8% | ~0.8% |
| Pulse Discharge | Room Temp | 0.78% | 0.07% |
| Pulse Discharge | 10°C | 1.01% | 0.04% |
| EUDC Discharge | Room Temp | 1.71% | 1.11% |
| EUDC Discharge | 10°C | 1.89% | 0.88% |
The mathematical foundation of the improved method explains its robustness. By solving Fick’s diffusion equation, the model internally tracks the lithium concentration \(c_s(r,t)\). The available capacity \(C_{ava}(t)\) is directly linked to the minimum lithium concentration within the particle, which evolves smoothly due to the parabolic nature of the diffusion equation. In contrast, the traditional method’s empirical capacity \(C_{total}(t)\) is an instantaneous function of the measured current \(I(t)\) and temperature \(T(t)\). A step change in current causes an immediate step change in the calculated capacity, leading directly to a discontinuity in the SOC derivative (\(dS/dt\)), manifesting as a jump in the SOC value. For a li ion battery management system (BMS), such jumps are undesirable as they can trigger erratic control actions.
Implementing the diffusion model solver in LabVIEW for online use required careful numerical consideration. The partial differential equation (PDE) was discretized using a central finite difference method in the spatial domain (\(r\)), transforming it into a system of ordinary differential equations (ODEs) in time. This system was solved efficiently using a built-in ODE solver within a MATLAB Script Node in LabVIEW, ensuring real-time feasibility. The computational overhead was manageable, as the single-particle model for a li ion battery strikes an excellent balance between physical fidelity and computational simplicity compared to full-order electrochemical models.
The role of temperature cannot be overstated in the context of li ion battery modeling. The Arrhenius correction for the diffusion coefficient \(D_s\) was instrumental in maintaining accuracy across the 10°C to 30°C range tested. At lower temperatures, lithium diffusion slows significantly, increasing concentration polarization and reducing effective capacity. The improved method inherently captures this through the temperature-dependent \(D_s\) in the diffusion equation. The traditional method attempts to capture this via the temperature term in its empirical fit, but our results show it is less precise, especially when extrapolating or under combined dynamic current and temperature swings.
Looking beyond the specific tests conducted, the implications of this work are significant for electric vehicle BMS design. The improved Ah-integral method offers a clear path to more reliable range prediction and battery protection. It is particularly advantageous for li ion battery packs experiencing the stop-and-go traffic of urban driving or the sustained high-power demands of highway acceleration, where current transients are frequent. Furthermore, the model’s foundation in electrode-level physics makes it potentially adaptable to different li ion battery chemistries (like Lithium Iron Phosphate or high-nickel NCM) by adjusting parameters such as the equilibrium potential curves and diffusion coefficients.
In conclusion, through this investigation, I have demonstrated that augmenting the conventional ampere-hour integration technique with a single-particle lithium diffusion model substantially enhances the accuracy and smoothness of State of Charge estimation for li ion batteries. The method successfully addresses the key shortcomings of the traditional approach: sensitivity to current and temperature variations, and the generation of unphysical SOC discontinuities during load transients. By directly linking the measurable electrical output of the li ion battery to the internal lithium concentration dynamics, this improved estimator provides a more truthful representation of the battery’s energy state. The implementation within a LabVIEW-based online monitoring system proves its practicality for real-time BMS applications. Future work will focus on extending this framework to include aging effects on model parameters and integrating it with state estimation filters like Kalman filters for further robustness against measurement noise. Ultimately, such physics-informed approaches are essential for unlocking the full performance, safety, and lifespan potential of every li ion battery in electric mobility and energy storage systems.